Quantifying AI return on investment is genuinely difficult because benefits often extend well beyond direct cost savings into productivity, innovation and decision quality — areas that resist a single tidy metric, but still need to be measured with discipline.
Where this gets hard
- Difficulty quantifying AI return on investment leads many organisations to avoid measurement altogether.
- Benefits often extend beyond direct cost savings into harder-to-measure areas like quality and speed of decisions.
- Measuring productivity, innovation and decision quality improvements requires different tools than measuring cost reduction.
- Determining appropriate performance metrics for each use case takes real effort that's frequently skipped in the rush to launch.
- Without agreed metrics, AI value ends up being argued anecdotally rather than demonstrated.
Where to start
- Define the appropriate metric type — cost, productivity, quality, innovation — for each use case individually, rather than forcing a single ROI formula.
- Establish a baseline before launch so improvement can actually be measured, not estimated after the fact.
- Use a mix of leading indicators (adoption, usage quality) and lagging indicators (business outcomes) for a fuller picture.
- Report AI value in the same forums and with the same rigour as other capital investment, not as a separate, softer category.
- Accept that some value (e.g. improved decision quality) may require qualitative alongside quantitative measures.
The consulting document includes a measurement framework template and a baseline-setting checklist for new AI initiatives.
For your top AI initiative, do you have a 'before' number to compare the 'after' against?